1{2 "document_id": "lb_qasper_ecb62b0ffa918a2faef02edcbcab0090da78c2c6e92612d1",3 "family": "public_qasper",4 "split": "test",5 "document_hash": "2cd25be0de1c6268c2e08ffc349de744fd71e8df7bfe0cb433507a31d65b8556",6 "questions": [7 {8 "question_id": "q00",9 "qtype": "qasper",10 "question": "What are the baselines?",11 "answer": "Vanilla ST baseline, encoder pre-training, in which the ST encoder is initialized from an ASR model, decoder pre-training, in which the ST decoder is initialized from an MT model, encoder-decoder pre-training, where both the encoder and decoder are pre-trained, many-to-many multi-task model where the encoders and decoders are derived from pre-trained ASR and MT models, Triangle+pre-train: BIBREF18 DBLP:conf/naacl/AnastasopoulosC18 proposed a triangle multi-task strategy for speech translation",12 "aliases": [13 "Vanilla ST baseline, Pre-training baselines, Multi-task baselines, Many-to-many+pre-training, Triangle+pre-train",14 "Vanilla ST baseline: The vanilla ST BIBREF9 has only a speech encoder and a decoder. It is trained from scratch on the ST-TED corpus.\n\nPre-training baselines: We conduct three pre-training baseline experiments: 1) encoder pre-training, in which the ST encoder is initialized from an ASR model; 2) decoder pre-training, in which the ST decoder is initialized from an MT model; and 3) encoder-decoder pre-training, where both the encoder and decoder are pre-trained. The ASR model has the same architecture with vanilla ST model, trained on the mixture of ST-TED and TED-LIUM2 corpus. The MT model has a text encoder and decoder with the same architecture of which in TCEN. It is first trained on WMT data (out-of-domain) and then fine-tuned on in-domain data.\n\nMulti-task baselines: We also conduct three multi-task baseline experiments including one-to-many setting, many-to-one setting, and many-to-many setting. In the first two settings, we train the model with $\\alpha _{st}=0.75$ while $\\alpha _{asr}=0.25$ or $\\alpha _{mt}=0.25$. For many-to-many setting, we use $\\alpha _{st}=0.6, \\alpha _{asr}=0.2$ and $\\alpha _{mt}=0.2$.. For MT task, we use only in-domain data.\n\nMany-to-many+pre-training: We train a many-to-many multi-task model where the encoders and decoders are derived from pre-trained ASR and MT models. "15 ],16 "answer_type": "qa_f1",17 "tolerance": 0.0,18 "supporting_fact_ids": [],19 "answer_position_frac": -1.020 }21 ],22 "facts": []23}